Personalized medicine in cancer treatment is based on knowledge of the RNA levels of a set of genes that are deregulated as a consequence of the pathology. Hence, knowledge of the gene expression profiles of tumours in cancer patients allows optimal treatments to be applied. One of the main challenges when classifying patients based on gene expression profiles is, precisely, finding genes whose expression is modified due to a certain disease. The identification of these genes allow patients to be classified into different groups that may present, for example, different responses to the same treatment. In this sense, those genes that have a high expression with respect to a healthy person for certain patients and a low expression for others are clear candidates to become prognostic markers of the pathology. For this reason, within the field of personalized medicine, it is of special interest to identify which genes present two groups of expression (high/low) depending on the patients. The identification of the number of groups with different gene expressions given certain data can be translated into the statistical problem of determining the number of modes (relative maxima of the density function) of the underlying distribution of the observed sample. In order to identify the number of groups, a statistical procedure based on the identification of the number of modes is presented. This method is applied to a gene expression dataset from breast cancer patients.